https://www.tensorflow.org/beta/tutorials/text/transformer https://github.com/tensorflow/models/tree/master/official/transformer/v2
Updated for TensorFlow 2, the old version is on the 1_x branch.
This is implemented in the most frictionless approach I have found so far for training keras models that save to SavedModels that can be used elsewhere. See 1_x branch for other approaches.
Demos for classification and translation from trained SavedModels are available.